CN105849643B - Yields estimation and control - Google Patents

Abstract

A kind of failure prediction methods that computer executes, for device manufacturing processes, the device manufacturing processes are related to the production substrate handled by lithographic equipment, the described method includes: carry out train classification models using training group, the training group include the measured value of procedure parameter associated with the production substrate as handled by the device manufacturing processes or determine value and about with the handled existing instruction for producing the associated defect of substrate in the device manufacturing processes under the value of the processing parameter；Output is generated with from disaggregated model, indicates the prediction of the defect for substrate.

Description

Yields estimation and control

Cross reference to related applications

This application claims the equity for the U.S. Provisional Application 61/917,305 submitted on December 17th, 2013, lead to herein It crosses to refer to and be incorporated by.

Technical field

This application involves lithographic equipments and process, good to improve for predicting and correcting defect more particularly, to one kind The tool of product rate.

Background technique

Lithographic equipment can be used in such as integrated circuit (IC) or the manufacture of other devices.In this case, scheme Case forms the circuit pattern (" design layout ") for the single layer that device (such as mask) be may include or be provided corresponding to device, and And this circuit pattern can radiate the methods of target part by the circuit pattern such as on patterning device, be turned Move on to be already coated on the substrate (such as silicon wafer) of radiation-sensitive materials (" resist ") layer target part (for example including One or more tube cores) on.In general, single substrate include be photo-etched equipment continuously, one at a time target part will be electric Road pattern is transferred to multiple adjacent target portions thereon.In a type of lithographic equipment, on entire patterning device Circuit pattern through once being transferred on a target part, such equipment is commonly referred to as wafer steppers.In one kind In the equipment (commonly referred to as stepping-scanning device) of substitution, projected bundle is along given reference direction (" scanning " direction) in pattern It is formed on device and is scanned, while along the direction synchronizing moving substrate parallel or antiparallel with the reference direction.Pattern forms dress The different piece for the circuit pattern set progressively is transferred on a target part.

Before circuit pattern is transferred to the element manufacturing process of substrate from patterning device in device manufacturing processes, Substrate may undergo the various element manufacturing processes of device manufacturing processes, such as linging, resist coating and soft bake.It is exposing After light, substrate bakes (PEB) after may undergoing the other devices production process of device manufacturing processes, such as exposure, develops, is hard It bakes.This series of element manufacturing process is used as the basis of the single layer of manufacture device (such as IC).Back substrate can The various element manufacturing processes of device manufacturing processes, such as etching, ion implanting (doping), metallization, oxidation, chemistry can be undergone Mechanical polishing etc., all these processes contribute to the single layer for being finally completed device.If needing multiple layers in device, that It will be for each layer of repetition whole process or its deformation.Finally, device will be arranged in each target part on substrate.Such as There are multiple devices for fruit, then these devices can be separated from each other by the technologies such as scribing or cutting, accordingly independent device Part may be mounted on carrier, be connected to pin etc..

Summary of the invention

A kind of failure prediction methods executed this application discloses computer, for device manufacturing processes, the device Manufacturing process is related to the production substrate handled by lithographic equipment, which comprises

Carry out train classification models using training group, the training group includes and the life as handled by the device manufacturing processes Produce the associated procedure parameter of substrate measured value or determine value and about under the value of the processing parameter in the device The existing instruction of the handled production associated defect of substrate in part manufacturing process；With

Output is generated from disaggregated model, indicates the prediction of the defect for substrate.

This application discloses a kind of methods of train classification models, which comprises

The defect in the substrate or on substrate is predicted using disaggregated model, and the disaggregated model has as independent variable The procedure parameter of device manufacturing processes for photolithographic exposure substrate and/or lithographic equipment will be used to be provided to the figure on substrate The layout parameter of case；

Receive the existing letter about the measured value for being directed to procedure parameter and/or layout parameter or the defect for determining value Breath；With

Based on the defect predicted and about the measured value for being directed to procedure parameter and/or layout parameter or determine lacking for value Sunken existing information carrys out train classification models.

This application discloses a kind of performed by computer, generation disaggregated models to promote lacking in device manufacturing processes The method for falling into prediction, the device manufacturing processes are related to the production substrate as handled by lithographic equipment, and the method includes using Training group carrys out train classification models, and the training group includes the process ginseng of multiple substrates as handled by the device manufacturing processes Several measured values determines value and the existing instruction about defect associated with the value of the processing parameter.

Disclosed herein as well is a kind of computer program product, including computer-readable medium, computer-readable Jie Matter has the instruction of record thereon, and described instruction realizes above-mentioned method when being executed by computer.

Detailed description of the invention

Fig. 1 is the block diagram of each subsystem of lithography system.

The method that Fig. 2 schematically shows defect in prediction device manufacturing processes.

Fig. 3 is the block diagram of simulation model.

Fig. 4 schematically shows the predictions of the processing window of layout.

The method that Fig. 5 schematically shows defect in the prediction device manufacturing processes according to an embodiment.

The method that Fig. 6 schematically shows re -training disaggregated model.

Fig. 7 shows the illustrative disaggregated model trained by training group.

Fig. 8 is exemplary the block diagram of computer system.

Fig. 9 is the block diagram of model predictive control system.

Figure 10 is the schematic diagram of lithographic projection apparatus.

Figure 11 is the schematic diagram of another lithographic projection apparatus.

Figure 12 is the more detailed view of equipment as shown in figure 11.

Figure 13 schematically shows the embodiment of lithographic cell or cluster.

Specific embodiment

Although specific with reference to the manufacture for being used for IC herein, it should be apparent that ground understands that the explanation of this paper can have Many other possible applications.For example, can be used for integrated optics system, the guidance of magnetic domain memory and detection pattern, liquid crystal Show the manufacture of panel, film magnetic head etc..It will be understood by those skilled in the art that in the case where this alternate application, it can With by any term " mask ", " chip " or " tube core " used herein think respectively with more upper term " mask ", " substrate " or " target part " is mutually general.

Herein, term " radiation " and " beam " are used to include various types of electromagnetic radiation, including ultraviolet radioactive (such as Wavelength with 365,248,193,157 or 126nm) and EUV (extreme ultraviolet radiation, such as with the wave within the scope of 5-20nm It is long).

Term " optimizing " as employed herein and " optimization " mean adjustment equipment, such as lithographic projection apparatus, So that the result and/or process (such as result and/or process of photoetching) of element manufacturing are with one or more more ideal Characteristic, the higher projection accuracy of the design layout on substrate, bigger processing window etc..

As A brief introduction, Fig. 1 shows illustrative lithographic projection apparatus 10A.Main component includes irradiation optics dress It sets, which defines partial ocoherence (being labeled as σ) and may include Optical devices 14A, 16Aa and 16Ab, to from radiation source The radiation of 12A shapes, and radiation source 12A can be deep knowledge base source or other including the source extreme ultraviolet (EUV) The source (as described above, lithographic projection apparatus itself is not needed with radiation source) of type；And Optical devices 16Ac, by pattern It is formed in the image projection to substrate plane 22A of the patterning device pattern of device 18A.In the pupil plane of projecting optical device The adjustable optical filter or aperture 20A at place can limit the range for the beam angle being mapped on substrate plane 22A, and maximum can The angle of energy defines the numerical aperture NA=sin (Θ of projecting optical device_max)。

In lithographic projection apparatus, the irradiation from source is guided and is shaped by patterning device by projecting optical device Onto substrate.Term " projecting optical device " be, is broadly defined here be include can change radiation beam wavefront it is any Optical component.For example, projecting optical device may include component 14A, 16Aa, at least some of 16Ab and 16Ac component.It is empty Between image (AI) be radiation intensity distribution at substrate level.Resist layer on substrate is exposed, and spatial image It is transferred to resist layer, as potential " Resist images " (RI) therein.Resist images (RI) can be defined as The spatial distribution of the solubility of resist in resist layer.Resist model can be used for calculating resist figure from spatial image Picture, example can be found in the U.S. Patent application of Publication No. US2009-0157630, and disclosure of the documents are logical Reference is crossed to be incorporated herein entire contents.Resist model only with the property of resist layer (such as exposure, exposure after roast The effect of roasting (PEB) and the chemical process occurred during development) it is related.Optical property (such as the source, pattern of lithographic projection apparatus Form the property of device and projecting optical device) it defines spatial image and can be defined in optical model.Because can be with Change the patterning device used in lithographic projection apparatus, it is desirable to by the optical property of patterning device with include The optical property of at least rest part of the lithographic projection apparatus in source and projecting optical device separates.

As shown in figure 13, lithographic equipment LA can form lithographic cell LC (otherwise referred to as photoetching member or photolithographic cluster) A part, lithographic cell LC further includes executed before one or more exposures on substrate and the equipment of post-exposure processes. Under normal conditions, these include to deposit one or more spinner SC of resist layer, to against corrosion after exposure One or more developer DE, one or more chill plate CH and one or more baking plates BK of agent development.Substrate Manipulation device or robot RO are from I/O port I/O1, I/O2 pickup substrate, then by it between different processing equipments It is mobile, and it is delivered to the loading bay LB of lithographic equipment.These devices for being often commonly referred to as track are in orbits controlling unit Under the control of TCU, the orbits controlling unit TCU itself is controlled by management control system SCS, the management control system SCS also controls lithographic equipment via photoetching control unit LACU.Therefore, different equipment can be operated to productivity and Treatment effeciency maximizes.Lithographic cell LC can also include for etching one or more etchers of substrate and being configured to survey Measure one or more measuring devices of the parameter of substrate.The measuring device may include being configured to the physical parameter of measurement substrate Optical measuring device, such as scatterometer, scanning electron microscope etc..Measuring device may be embodied in lithographic equipment LA. The embodiment of the present invention can be realized in management control system SCS and/or photoetching control unit LACU or by management control System SCS and/or photoetching control unit LACU is realized.For example, coming from management control system SCS and/or photoetching control unit The data of LACU can embodiment through the invention use, one or more signals from the embodiment of the present invention can be with It is provided to management control system SCS and/or photoetching control unit LACU.

The method that Fig. 2 schematically shows the defects of prediction device manufacturing processes.Defect can be systematic defect, Such as constriction (necking), line retract (line pull back), line narrow, CD, overlap joint and bridge joint；Defect be also possible to Machine defect, such as the defect caused by particle (such as dust granule) deposition.Systematic defect can be predicted and control. Defect can Resist images, optical imagery or etching pattern (i.e. by using the resist on substrate layer as mask come It is etched and is transferred to the pattern on substrate layer) in.Computation model or empirical model 213 can be used for predicting defect 214 (such as the prediction presence of defect, the position of defect, the type of defect, shape of defect etc.).Model 213 is it is contemplated that device system Make the parameter 211 (also referred to as procedure parameter) and/or layout parameter 212 of process.Procedure parameter 211 is and device manufacturing processes It is associated and the parameter unrelated to layout.For example, procedure parameter 211 may include characteristic (such as the intensity, pupil wheel in source Exterior feature etc.), the characteristic of projecting optical device, dosage, focusing, the characteristic of resist, the characteristic of resist development, anti-aging drug The characteristic and/or etching characteristic baked afterwards.Layout parameter 212 may include the shape of various features in layout, size, opposite The overlapping of feature on position and absolute position, and different layouts.Model 213 can be fixed model, i.e. model itself (such as procedure parameter 211 and layout parameter 212) is not inputted with it and change.That is, under same input, fixed The result of model is always identical.In empirical model, image (such as Resist images, optical imagery, etching pattern) is not It is modeled；And empirical model predicts defect based on the correlation between input and defect.In computation model, the one of image Part or characteristic are calculated, and defect is distinguished based on the part or characteristic.For example, line rollback defect can pass through It was found that leave its desired position too far and identified for thread end；Bridge defects can be by finding that two lines undesirably connect Location and it is identified.

Fig. 3 shows illustrative computation model.Source model 31 indicates optical characteristics (including the radiation intensity distribution in source And/or phase distribution).Projecting optical device model 32 indicates that the optical characteristics of projecting optical device (including is filled by projection optics Set the variation of caused radiation intensity distribution and/or phase distribution).Design layout model 35 indicates that the optics of design layout is special Property (variation including radiation intensity distribution and/or phase distribution as caused by given design layout), be that pattern is formed The expression of the arrangement of the feature formed on device or by patterning device.Spatial image 36 can be by source model 31, projected light It learns mounted cast 32 and design layout model 35 is simulated.Resist and/or etching can be used in resist and/or etching pattern 38 Model 37 is simulated by spatial image 36.Profile and CD in the picture can be predicted for example to the simulation of photoetching.

More specifically, it is noted that source model 31 can indicate the optical characteristics in source, and including but not limited to Sigma (σ) sets Fixed and any specific irradiation source shape (such as annular, quadrupole and bipolar etc. off axis radiation source etc.).Projected light The optical characteristics of projecting optical device can be indicated by learning mounted cast 32 comprising aberration, deformation, refractive index, physics size, object Manage size etc..Design layout model 35 also may indicate that physical patterns form the physical property of device, as described, such as United States Patent (USP) No.7 described in 587,704, is hereby incorporated by reference in its entirety..The target of simulation is essence Really predict positioning, aerial image intensity slope and the CD at such as edge, it later can be compared with desired design.It is described Desired design is normally defined preparatory OPC design layout, may be provided as standard digital file format (such as GDSII Or OASIS) or other file formats.

The processing window that Fig. 4 schematically shows layout (there is no mistake at the layout of Systematic Errors The space of journey parameter) prediction.Subprocess window 421-423 (as shown in the region for not beating hacures) can be directed in layout Feature using model (empirical model or computation model) relative to layout 411-413 different types of defect (for example, line return Move back, CD, constriction etc.) it is predicted.It retracts for example, not generating line in these features in the process in subprocess window 421 Defect.All features and for all types of defects subprocess windows can be combined to form the process window of layout Mouth 430.

The method that Fig. 5 schematically shows defect in the prediction device manufacturing processes according to an embodiment.Disaggregated model ( Referred to as classifier) 513 can be used to predict defect 514 (such as prediction the presence of defect, the position of defect, defect class Type, shape of defect etc.).Model 513 is it is contemplated that procedure parameter 511 and/or layout parameter 512.Procedure parameter 511 is and device Part manufacturing process is associated and the parameter unrelated to layout.For example, procedure parameter 511 may include source characteristic (such as Intensity, pupil profile etc.), the characteristic of projecting optical device, dosage, focusing, the characteristic of resist, resist development spy The characteristic and/or etching characteristic baked after property, anti-aging drug.Layout parameter 512 may include the various features in layout The overlapping of feature on shape, size, relative position and absolute position, and different layouts.

Term " classifier " or " disaggregated model " indicate mathematical function sometimes, which is realized by sorting algorithm, right Input data is classified.In machine learning and statistics, classification is based on including the known observation in category membership (observation) training group of the data of (or example) distinguishes the problem of which classification group (subgroup) new observation belong to. Each observation is resolvable to one group of quantifiable property, which is known as various explanatory variables, feature etc..These Property can carry out various classification, and (for example, " good ", --- not generating process or " poor " of defect --- generates the mistake of defect Journey).The example that classification is considered to be supervised learning, that is, training group of the study by the observation correctly distinguished is in what situation Under be achievable.

There are many cross-cutting term variation.In statistics, (logistic can be returned by Luo Jisi in classification Regression) or similar procedure is come in the case where realizing, the property of observation be referred to as explanatory variable (or independent variable, Regression variable (regressor) etc.), the classification being predicted is known as a result, it is considered to be the possible of non-independent variable Value.In machine learning, which is commonly referred to as example, and explanatory variable is referred to as feature (being grouped into characteristic vector), It is class by the possible classification being predicted.

Disaggregated model can be expressed by linear function, the linear function by using dot product by the characteristic vector of example with Weight vectors are combined so that score is dispatched to each possible classification k.The classification predicted is that have the classification of top score.It should The scoring function of type is referred to as linear predictor function and has following general type: score (X_i, k) and=β_k·X_i.Wherein X_iIt is the characteristic vector of example i, β_kCorrespond to the weight vectors of classification k, score (X_i, k) and it is to be dispatched to classification with by example i The associated score of k.Model with the basic boom is referred to as linear classifier.The example of this algorithm is that Luo Jisi is returned Return, multinomial Luo Ji (multinomial logit), probability unit recurrence, perceptron algorithm, support vector machine (support Vector machine), input vector machine (import vector machine) and/or linear discriminent analyze (linear discrminant analysis)。

In one embodiment, disaggregated model 513 is related to Luo Jisi recurrence.In the case of the embodiment, non-independent variable It is binary, that is, the quantity of available classification is two, such as " good " or " poor ".However, the quantity of available classification is certainly not It is limited to two.

Luo Jisi returns the non-independent variable measured as the predicted value of non-independent variable by using probability in classification With the relationship between one or more independent variables, usually (but being not required) can be with for wherein one or more independent variables It is continuous.The disaggregated model 513 can be used including-kind or more process and/or layout parameter data training group with And the process and/or layout parameter are to generate defect (i.e. " poor ") not generating defect (" ") to be trained.Initial Training group can be obtained from one or more test runs of the layout under a certain range of parameter value.

In one embodiment, disaggregated model 513 is related to kernel Luo Jisi recurrence, especially when scoring function cannot be with linear Form score (X_i, k) and=β_k·X_iCome when indicating, wherein X_iIt is the characteristic vector of example i, β_kCorrespond to the weighting arrow of classification k Amount, score (X_i, k) and it is to be dispatched to the associated score of classification k with by example i.Kernel can be initially used for independent variable (such as procedure parameter) projects in another parameter space: Φ: X → Y, so that score (X_i, k) and=β_k·Y_i, wherein Y_i =Φ (X_i)。

Method shown in fig. 5 can also include aligning step 515, wherein adjustable one or more procedure parameters 511, one or more layout parameters 512 or the two are to reduce or eliminate defect.

In one embodiment, model 513 is not fixed model.Alternatively, model 513 can be according to the number from measurement According to 516, yields data (such as passing through the measuring tools such as such as electron microscope, the discrimination by Electronic Testing etc. to defect) or User from lithographic equipment or other data from alternate model (such as another empirical model or computation model) carry out It is perfect.Model 512 can use other data after one or more tube cores and/or one or more substrate exposures It carries out perfect.

Data 516 may include the survey of procedure parameter associated with the multiple production substrates handled by device manufacturing processes Magnitude determines value.Production substrate is the substrate in one or more production phases with one or more devices.Example Such as, production substrate can be the substrate with the Resist images for being directed to one or more devices.For these substrates The value of procedure parameter may include the data (such as equipment setting and/or lithographic equipment sensing data) from lithographic equipment And/or metric data (such as being provided by dedicated optical measuring device to measure the physical parameter of Resist images).As another One example, production substrate can be with the substrate through overetched feature and/or the substrate with the feature with function element. Value for the procedure parameter of these substrates may include from etch tool data (such as etch tool setting and/or Etch tool sensing data) and/or metric data (such as being provided by scanning electron microscope) and/or yields data (example Such as from the defect analysis of measuring tool, the device produced and desired device, the Electronic Testing of device etc. are carried out pair Than).In addition, the device manufacturing processes can be related to from substrate to device or part thereof of whole process.For example, the device system The process of making can be only lithographic patterning process or combine with another device fabrication.In one embodiment, device manufactures Process can be only etching process or combine with another device fabrication.In etching, which is related to And lithographic equipment, because the substrate handled by etching machines carries out pattern by being related to the lithographic patterning process of lithographic equipment Change.

In one embodiment, provide a kind of instruction existing for defect, the defect under procedure parameter value Processed production substrate is associated in device manufacturing processes.Therefore, in one embodiment, the measured value of procedure parameter or determination Each of value is associated with the instruction existing for defect.For example, the existing instruction about defect can be for table Show present or absent any mark of defect.For example, the mark can be " good " and/or " poor ".The mark can be by making User application or use can application tool automatically determine.For example, the Electronic Testing of substrate can be with the defects of recognition means With mark device " good " or " poor ".The substrate tested is associated with the value of procedure parameter.In one example, if the yields Under some threshold value, the value of associated procedure parameter 511 and/or layout parameter 512 can be labeled/be categorized into " poor ". The combination of the value and mark of procedure parameter (such as dosage and focusing) is used for training pattern 513.

Data from measurement can from Optical measuring tool (such as measure from measure target and/or from warp The tool of the diffraction radiation in the region of overexposure), electron microscope or other suitable checking tools obtain, and can be by Data or aligned data measured by sensor (such as horizon sensor) in lithographic equipment.

In one embodiment, the perfect of model 513 may include using including one or more procedure parameters or process The new observation of both parameter and layout parameter and data, yields data or the use from lithographic equipment from measurement The training group of family or other data from alternate model is trained, and wherein the procedure parameter and layout parameter are used in one Or more tube core and/or one or more substrates exposure.Training group for sophisticated model 513 may not necessarily include It is used for all data of training pattern 513 before.For example, if model 513 is carried out initially with the data group for including 100 observation Training, then the training group may include 99 and new observation in this 100 observations.This method may limit training group Size is to limit trained calculating cost.One or more algorithms can be used for management or the continuously ruler of management data set It is very little.For example, input vector machine or support vector machine can be used for the size of management data set.

A kind of method that Fig. 6 schematically shows train classification models (such as model 513).In step 611, device Defect, such as in the resist or the defects of the optical imagery of device, predicted using disaggregated model, wherein being used to form Independent variable of one or more processes and/or layout parameter of device as disaggregated model.In step 612, the device, Such as resist or optical imagery are modeled or using suitable inspection using alternate model (such as empirical model or computation model) Tool is looked into measure, such as Resist images or the pattern being etched, and the presence of defect, shape, type and/or position are true It is fixed.In step 613, which is trained based on the prediction of the defect determined by simulating or measuring and presence.

Fig. 7 shows the exemplary output for the disaggregated model trained by training group, and the training group includes 187 observations, It includes as procedure parameter focusing (vertical axis) and dosage (horizontal axis) to whether it is described focusing or dosage to production (" O " indicates no defect, such as " good " to raw defect；" X " indicates defect, such as " poor ").The probability of defect leads to after training It crosses model determination and is shown as isogram in the output of Fig. 7, wherein probability is associated with isopleth.It should be appreciated that should The output of model can be other forms, such as colored or tonal gradation or result list.The output of the model with should Training group matches very well with reasonable manner.

Therefore, in one embodiment, provide and be related to the on-line study of the device manufacturing processes of lithographic equipment.That is, Make disaggregated model, the disaggregated model be based on one or more procedure parameters associated with production substrate (such as dosage and/ Or focus) new measured value or determine that value and the instruction of defect associated with this one or more procedure parameter continue Ground is regularly trained to.Therefore, in one embodiment, it is exclusively used in the model quilt of lithographic equipment and/or device manufacturing processes Production, with developing together with the use of time and lithographic equipment and/or the processing using the substrate of device manufacturing processes.Cause This, in one embodiment, empirical data indicates device manufacturing processes (for example, using specific patterning device for generating Layout specific device manufacturing processes) model and then to the model progress shaping.

Due to empirical data, data explanation may be hardly needed.For example, this process can in the training disaggregated model Directly to use the pupil intensity distribution map of measuring tool, rather than this intensity distribution is construed to appropriate structure and side The data improved model is used after edge slope.For example, learning art can be had defect based on the model for having already passed through study Probability it is related to the pupil intensity distribution map, therefore, the data from pupil intensity distribution map can help to confirm that defect is deposited (or being not present).Therefore, in one embodiment, this data do not need itself associated mark, but can use In increasing the model prediction, whether defect is by the ability of appearance.It is being had for example, the pupil intensity distribution map can indicate to have The feature of parameter value except the line of adjacent parameter value.Although not necessarily can confirm that defect, this information can be used for The training model is to help to confirm or deny the existing correlation about this feature and shortage probability in a model.Implement one Example in, measurement be using such as Optical measuring tool measurement pupil device layout.Therefore, specific structure (such as is measured Survey target) it may not be required；Any structure is all possible, such as device layout structure (such as in logic/MPU device In sram cell block).

In one embodiment, knowledge/experience of operator may be used as the input to the structure of sorter model.Operator Feedback can manipulate the predictive ability of the model.For example, the user can add more predictive features into sorting algorithm.

In one embodiment, which can be generated the specific measurement of given defect or determines the general of feature Rate.Moreover, the precision of the prediction increases with time and more measurement because sorter model can become more " have through It tests ".The on-line study is different with so-called data mining, and being commonly used for check, why thing can malfunction.It is real one It applies in example, online data can predict the probability of defect occur when processing is carrying out for generating and then updating.Therefore, The output of the model can provide indicator that can be checked and which substrate tube core to measure it is all to see whether Relationship is all good, to improve whole yields.

In one embodiment, which can be used sorter model to control.The on-line study allows pair The process carries out tracking (such as drift) and allows adjustment/fine tuning (control) of the process.For example, one of lithographic equipment or more Multiple parameters can the output based on sorter model and controlled, regardless of be it is automatic or user estimation after. For example, the focusing of lithographic projection apparatus and/or dosage can be controlled based on output.

In one embodiment, sorter model includes the measurement (such as being not only the measurement of data in field) across substrate. Therefore, which can be the data by focusing, because local substrate difference may be relatively large.Therefore, by depending on Measurement from multiple and different substrates, can be independent without generating with the focusing that specific dosage measurement passes through completely Exposure.

In one embodiment, although the discussion has focused on photolithographic parameters (such as focusing and dosage), study Example can be easily extended to other processes, such as etch.For example, relationship between etch features and yields can be by Study can also be observed with online measuring data.

Therefore, in one embodiment, a kind of Study strategies and methods model that can be predicted defect and estimate its probability is provided. In addition, in one embodiment, which is not static and during device manufacturing processes by measuring or determining Data are continuously updated and improve.In addition, the sorter model can be by will be with the incoherent data of photolithography (such as thickness Data, operator's determining defects etc. after variation, etching) model is fed to be extended in its coverage, this further increases Strong " experience " of the model.

In one embodiment, which can be directed to unmeasured data point to improve estimating for shortage probability Meter.For example, can prediction to the defect at the position A and B on substrate or on pattern layout it is interested.The sorter model energy Enough appearance of the prediction defect at position A and B.Then, increasing associated with position A information, (such as metric data (has and marks Know data or do not have mark data), yields data etc.) measurement can be used for further training the sorter model. Here, the estimation of probability of the defect at position A and B can be determined, without measuring at the B of position.

In one embodiment, which may include various types of information.Thus, for example, the classification Device model may include about the data for being directed to specific dosage with focusing combined shortage probability, further include about which Or more equipment is associated with the data point, which pattern layout is associated with the data point, which etching type used Etc. data.Therefore, in one embodiment, sorter model can be for (such as the only agent of limited and specific data Amount and focus information and associated mark) Lai Xunlian or some variants for comprehensive data or therebetween are trained.Institute Can be defined from the model for focusing on particular device (such as lithographic equipment), specified arrangement etc. from more fully model " submodel ".Therefore, user can use specific model or " submodel ", no matter for example for analysis or for excessively program-controlled System is not always the case, this is determined according to the needs or requirement that focus on user.

In one embodiment, the training of sorter model decides whether that new training data (such as data point of measurement) adds Add enough information to be included in model.It adds the new information and is increased the size of the model and balanced, so that the mould Type will not be grown to non-boundary.

In one embodiment, sorter model can provide the prediction to the impressionability on product.For example, classifier mould Type can quantify the probability of defect.Sorter model can provide the prediction of full lining bottom.Sorter model can predict the number of defect Amount.Sorter model can predict the yields of good tube core.Sorter model can provide the position of defect.

In one embodiment, the training data for being directed to sorter model can be directed to be produced by device manufacturing processes Batch in each substrate sampled.The training device data can be directed to substrate (such as substrate in a collection of substrate) On each device sampled.The training data can be directed to each layer on substrate and be sampled.In one embodiment, The training data includes measurement data, and measurement position can in tube core, from die external (such as scribing line measures target) and/ Or across substrate.Measurement in tube core can analog result based on hot spot, specific device architecture position (such as SRAM And other positions) sample and/or be to rely on device (such as IC).

In one embodiment, new training data be continuously supplied (for example, in device manufacturing processes, at multiple batches During secondary substrate etc.) and therefore predict that quality is constantly updated.

In one embodiment, the output of sorter model can be provided to the yields management system of manufactory to improve Device yield.

Therefore, in one embodiment, the value of one or more procedure parameters and/or layout parameter by machine learning with Yields sensitivity statistical correlation on the product of entire substrate.Thus, for example, defect can be directed to the specific part of substrate And/or the specific part of tube core is predicted.Further, which can be directed to actual production based on actual production data come pre- It surveys.Then, machine learning model can be directed to actual process conditions (including such as their drift) and can be realized than making The stronger prediction of prediction for being based only on theoretical model with pattern layout data.

Therefore, in one embodiment, a kind of comprehensive defect inspection and yield prediction system are provided, wherein online With the parameter value (such as from measuring tool for measuring product substrate) for product for inferring impressionability defect.It should System is using artificial intelligence technology come the impressionability of the defect for being directed to layout of forecasting system.This is by hotspot prediction from list A tube core expands to entire substrate, including substrate edge area.In one embodiment, which can be used measuring tool measurement Each substrate in one batch.The system can be enhanced or substitute current defect detecting method (such as scanning electron microscopy Mirror).The output of the system may include the performance indicator in the yields management system of manufactory to predict and/or improve The yields of resulting devices.The lithographic equipment of customization or the matingplan of other equipment and file can be generated with (certainly in the system The yields of the subsequent batches (or substrate) of for example same device and/or layer is improved dynamicly).The system can be realized for The continuous estimation of defect and tracking simultaneously can continuously improve model prediction accuracy.The appearance of defect can by adjusting (such as Closed-loop control) and be reduced or minimized.

The advantage of embodiment described herein may include faster ratio defective product slope, more effective SEM check, history point Analysis and/or control.

Fig. 8 is the block diagram according to the model predictive control system of an embodiment.As shown, one or more inputs 800 are provided to device manufacturing processes 810, which is related to carrying out patterned production using lithographic equipment Substrate.Input 800 may include one or more layout parameters as described above and/or one or more procedure parameters. The device manufacturing processes 810 are related at least one devices productive steps, such as lithographic patterning, development, etching etc. or therefrom Any combination of the step of selection.

After or during the period process 810, it can produce one or more outputs 820.The output 820 may include making One or more processes of substrate and/or the value of layout parameter are produced made by the device manufacturing processes.For example, the value Can be as measured by measuring tool produce substrate data, can be from lithographic equipment or production substrate processing it Data of etch tool afterwards etc..In one embodiment, at least some of the data can be marked as described above.It is this Output 820 is provided to the model described with training of state estimator 830.In one embodiment, which is used to predict defect, Although model can be trained to predict other aspects.As shown, the state estimator 830 can receive for process 810 One or more inputs 820.For example, one or more inputs 820 can be topology data or be produced by topology data Raw data.For example, the data as caused by topology data can be the simulation of pattern layout to distinguish hot spot (for example, pattern The region of layout tends to improperly pattern).The simulation softward in this field can be used to make in this data being modeled Make, such as the Tachyon LMC product of ASML '.

Then the model of state estimator 830 may be used to provide to the output of adjuster 840.Adjuster 840 can mention For one or more inputs 800 to process 810 and/or modify one or more inputs of the process that is provided to 810 800.For example, one or more settings for lithographic equipment, etch tool etc. can be generated in the adjuster 840, to help Alleviate the defects of the following production in substrate.In one embodiment, adjuster 840 can receive one or more targets 850, the target 850 distinguishes that adjuster 840 should generate or modify one or more inputs for process 810 800 or adjuster 840 what standard should generate or modify one or more inputs 80 for process 810 with.

Fig. 9 is the illustrative block diagram for illustrating computer system 100, can assist realizing and/or executing public herein The optimization method and process opened.Computer system 100 includes: bus 102 or other communication mechanisms for information communication；With with The processor 104 (or multiple processors 104 and 105) for being used to handle information that bus 102 couples.Computer system 100 is also wrapped It includes main memory 106 (such as random access memory (RAM) or other device for dynamic storage), the main memory 106 couples To bus 102 for storing and/or providing the information and instruction executed by processor 104.Main memory 106 can be also used for Temporary variable or other average informations are stored and/or provided during the execution of the instruction executed by processor 104.Computer system 100 can also include the read-only memory (ROM) 108 or other static storage devices for being coupled to bus 102, be used to store And/or provide the static information for processor 104 and instruction.Storage device 110 (such as disk or CD) is provided and coupling It is connected to bus 102, for storing and/or providing information and instruction.

Computer system 100 can be coupled to display 112 (such as cathode-ray tube (CRT) or plate via bus 102 Or touch panel display), for showing information to computer user.Input unit 114 (including alphanumeric key and other Key) it can be coupled to bus 102, for information and command selection to be communicated with processor 104.Another type of user's input Device is cursor control 116 (such as mouse, trace ball or cursor direction key), for by directional information and command selection with Processor 104 communicates and for controlling the movement of the cursor on display 112.This input unit is typically in two axis ( One axis (such as x) and second axis (such as y)) on tool there are two freedom degree, this allows the position in described device given plane It sets.Touch panel (screen) display is also used as input unit.

According to one embodiment of present invention, a part of of optimization process can be by computer system 100 in response to being used for Execution includes the processor 104 of one or more sequences of one or more instructions in main memory 106 and is held Row.Such instruction can be read in main memory 106 from another computer-readable medium (such as storage device 110). It include the execution of the sequence of the instruction in main memory 106 so that processor 104 executes method and step described herein.More One or more processors in processing arrangement can also be used to carry out include instruction in main memory 106 sequence Column.In alternative embodiment, hard-wired circuitry can be used for substituting software instruction or in conjunction with software instruction.Therefore, exist This description carried out is not limited to any specific combination of hardware circuit and software.

Term " computer-readable medium " as employed herein indicates to participate in providing instructions to processor 104 to execute Any medium.Such medium can use many forms, including but not limited to non-volatile media, Volatile media and biography Defeated medium.Non-volatile media includes such as CD or disk, such as storage device 110.Volatile media includes dynamic memory Device, such as main memory 106.Transmission medium includes coaxial cable, copper conductor and optical fiber, includes the conducting wire of bus 102.It passes Defeated medium can also using the form of sound wave or light wave, such as generated during radio frequency (RF) and infrared (IR) data communication this A little sound waves or light wave.The usual form of computer-readable medium includes such as floppy disk, soft dish (flexible disk), hard disk, magnetic Band, any other magnetic medium, CD-ROM, DVD, any other optical medium, punched card, paper tape, its any with sectional hole patterns It is its physical medium, RAM, PROM and EPROM, FLASH-EPROM, any other memory chip or cassette tape, as described below Any other medium that carrier wave or computer can be read.

Various forms of computer-readable mediums may relate to one or more sequences of one or more instructions It is sent to processor 104, for executing.For example, instruction can initially appear on the disk or memory of remote computer.Far Instruction can be loaded into its dynamic memory and send described instruction on communication path by journey computer.Computer system 100 can receive the data from communication path and data are placed in bus 102.Bus 102 loads data to primary storage Device 106, processor 104 are obtained and are executed instruction from main memory 106.It can be may be selected by the received instruction of main memory 106 Ground is stored on storage device 110 before or after the execution of processor 104.

Computer system 100 may include being coupled to the communication interface 118 of bus 102.The offer of communication interface 118 is coupled to The bidirectional data communication of network link 120, the network link 120 are connected to network 122.For example, communication interface 118 can provide Wired or wireless data communication connection.In any such embodiment, communication interface 118 send and receive electricity, electromagnetism or Optical signal carries the digit data stream for indicating various types of information.

Typically, data communication is provided to other data sets by one or more networks by network link 120.Example Such as, network link 120 can provide the number for being connected to host 124 or being operated by Internet service provider (ISP) 126 by network 122 According to equipment.ISP126 provides data further through worldwide packet data communication network (being commonly referred to " internet " now) 128 and leads to Telecommunications services.Network 122 and internet 128 are all using the electricity for carrying digit data stream, electric magnetically or optically signal.Pass through various networks On signal and network link 120 and by the signal of communication interface 118 by numerical data transmission to computer system 100 and from Computer system 100 is sent back, these signals are the exemplary forms for transmitting the carrier wave of information.

Computer system 100 can send message by network, network link 120 and communication interface 118 and receive data, Including program code.In the example of internet, server 130 can pass through internet 128, ISP126, network 122 and communication Interface 118 is that application program sends request code.For example, the application program being downloaded as one can be provided for implementing this The code of literary the method.It is received at it and/or in storage device 110 or for other non-volatile storages of execution later When being stored in storage, received code can be executed with device 104 processed.In this way, computer system 100 can obtain At the application code of carrier format.

Figure 10 schematically shows illustrative lithographic projection apparatus.The equipment includes:

Irradiation system IL, for adjusting radiation beam B.In this specific situation, irradiation system further includes radiation source S O；

First objective table (such as mask platform) MT is provided with the figure that device MA (such as mask) is formed for holding pattern Case forms device retainer and is connected to the first positioning device PM, forms device to be accurately with respect to component PS registration pattern；

Second objective table (substrate table) WT is provided with the lining for keeping substrate W (such as silicon wafer of coating resist) Bottom retainer is simultaneously connected to the second positioning device PW, substrate is precisely located relative to component PS；

Optical projection system PS (such as refraction type, the optical system that reflective or refraction is reflective), by patterning device MA Exposure part be imaged on the target part C (for example including one or more tube cores) of substrate W.

As shown here, the equipment is transmission-type (having transmissive mask).However, for example, usually it goes back It can be reflective (there is reflection type mask).Alternatively, the equipment can form dress using another type of pattern It sets to substitute the use of classical mask；Example includes array of programmable mirrors or LCD matrix.

Source SO (such as mercury lamp or excimer laser) generates radiation beam.This is a branch of to be a supplied directly to irradiation system In (illuminator) IL, or it is being supplied in irradiation system (illuminator) IL later across regulating device (such as beam expander).Irradiation Device IL may include adjustment device AD, and the intensity distribution that the adjustment device AD is used to be set in beam is externally and/or internally Radial extension (typically referred to as σ-outside and σ-inside).In addition, it generally includes various other components, such as polymerizer IN With condenser CO.In this way, the beam B being irradiated on patterning device MA has desired uniformity and intensity in its cross section Distribution.

About Figure 10, it should be noted that source SO can be located in the shell of lithographic projection apparatus, (when source, SO is such as mercury Often such situation when lamp), but it can also be far from lithographic projection apparatus, the radiation beam generated is directed into described In equipment (such as with the help of suitable directional mirror)；The latter situation is usually that work as source SO be excimer laser Device (is e.g. based on KrF, ArF or F₂The excimer laser of laser) situation.

The patterning device MA that beam B is then retained on patterning device platform MT is intercepted.Have already passed through pattern It is formed after device MA, the beam B passes through optical projection system PS, and beam B is focused on the target part C of substrate W.It is fixed second Under the auxiliary of position device PW (and interfering meter measuring device IF), substrate table WT can be moved accurately, such as so as on the road of beam B Different target part C is positioned on diameter.Similarly, such as mechanical from patterning device library patterning device MA is obtained Later or during scanning, the first positioning device PM can be used for that patterning device is precisely located relative to the path of beam B MA.In general, in the help of long-stroke module (coarse positioning) and short stroke module (fine positioning) (not shown clearly in Figure 10) Under, realize the movement of objective table MT, WT.

It can be used mask alignment mark M1, M2 and substrate alignment mark P1, P2 carry out alignment pattern and form device (such as to cover Mould) MA and substrate W.Although substrate alignment mark shown in occupies dedicated target portion, they can be located at target portion Divide in the space between (these are known as scribe-lane alignment marks).Similarly, it is arranged by more than one tube core in pattern shape At device (such as mask) MA it is upper in the case where, the mask alignment marks can be between the tube core.It is small Alignment mark may also be included in that in tube core, between device feature, in such circumstances it is desirable to which the label is as far as possible It is small and do not need any imaging or treatment conditions different from adjacent feature.

Figure 11 schematically shows another illustrative lithographic projection apparatus 1000.The lithographic projection apparatus 1000 wraps It includes:

Source collector module SO；

Irradiation system (illuminator) IL is configured to adjust radiation beam B (for example, EUV radiation)；

Support construction (such as mask platform) MT, being configured to support patterning device (such as mask or mask), MA is simultaneously It is connected with the first positioning device PM for being configured to be precisely located patterning device；

Substrate table (such as wafer station) WT, be configured to keep substrate (such as chip coated with resist) W, and with match The second positioning device PW for substrate to be precisely located is set to be connected；With

Optical projection system (such as reflective projection system) PS, the optical projection system PS is configured to will be by patterning device The pattern that MA assigns radiation beam B projects on the target part C (for example including one or more tube cores) of substrate W.

As shown here, the equipment 1000 is reflection-type (for example, using reflection type mask).It should be noted that by It is absorbability in EUV wavelength range in most of materials, therefore patterning device can have mattress reflector, including Such as more laminations of molybdenum and silicon.In one example, more lamination reflectors have 40 layers of pairs of molybdenum and silicon.Use X-ray lithography Art can produce even smaller wavelength.Since most of materials are absorbabilities in EUV and X-ray wavelength, so scheming Case forms the thin slice of the patterned absorbing material (for example, the TaN absorber on the top of mattress reflector) on device pattern Define that feature will print (positive corrosion-resisting agent) or not print the region of (negative resist).

Referring to Fig.1 1, illuminator IL receive extreme ultraviolet (EUV) radiation beam from source collector module SO.To generate The method of EUV radiation including but not necessarily limited to converts the material into plasmoid, which has and have in EUV range At least one element of one or more emission lines, such as xenon, lithium or tin.In commonly referred to as plasma generation with laser In a kind of such method of (" LPP "), which can irradiate fuel by using laser beam to generate, and fuel is for example It is drop, line or the cluster of the material with transmitting line element.Source collector module SO can be including being used for for providing Excite a part of the EUV radiation system of the laser (being not shown in Figure 11) of the laser beam of fuel.It is formed by plasma Body emits output radiation, such as EUV radiation, is collected by using the radiation collector being arranged in the collector module of source.Swash Light device and source collector module can be discrete entity, such as when using CO₂Laser provides laser beam and is used for fuel fired When be such.

In this case, laser is not considered as forming a part of lithographic equipment, also, by means of including for example closing The beam transmission system of suitable directional mirror and/or beam expander, radiation beam are transferred to source collector module from laser.At it In the case of him, the source can be the component part of source collector module, such as when source is that electric discharge generates plasma EUV generation It is such when device (the commonly referred to as source DPP).

Illuminator IL may include adjuster, be configured for the angular intensity distribution of adjustment radiation beam.In general, can be right At least described externally and/or internally radial extension of intensity distribution in the pupil plane of the illuminator is (typically referred to as σ-outside and σ-inside) it is adjusted.In addition, the illuminator IL may include various other components, such as the reflection of facet field Lens device and facet pupil reflector apparatus.The illuminator can be used to adjust the radiation beam, in its cross section With required uniformity and intensity distribution.

The radiation beam B is incident on the patterning device (example being maintained on support construction (for example, mask platform) MT Such as, mask) on MA, and pattern is formed by the patterning device.Via patterning device (for example, covering Mould) after MA reflection, for the radiation beam B by optical projection system PS, beam is focused on the mesh of the substrate W by the optical projection system PS It marks on the C of part.By the second positioning device PW and position sensor system PS2 (for example, interferometric device, linear encoder or Capacitance sensor) help, the substrate table WT can be moved, such as accurately so that different target part C to be positioned at In the path of the radiation beam B.It similarly, can be by the first positioning device PM and another position sensor system PS1 For patterning device (for example, mask) MA to be precisely located relative to the path of the radiation beam B.Pattern shape can be used Shape device alignment mark M1, M2 and substrate alignment mark P1, P2 carry out alignment pattern and form device (for example, mask) MA and substrate W.

Equipment shown in can inciting somebody to action is used at least one of following mode:

It is substantially static same remaining support construction (such as mask platform) MT and substrate table WT 1. in step mode When, the entire pattern for assigning the radiation beam is once projected on target part C (that is, single static exposure).Then will The substrate table WT is moved along X and/or Y-direction, allows to expose different target part C.

2. in scan pattern, (so-called along assigned direction to support construction (such as mask platform) MT and substrate table WT " scanning direction ") while be synchronously carried out scanning, the pattern for assigning the radiation beam is projected on target part C (that is, single One dynamic exposure).Substrate table WT can pass through the throwing relative to the speed of support construction (such as mask platform) MT and direction (diminution) magnifying power and image reversal characteristics of shadow system PS determines.

3. in another mode, will be used to keep support construction (such as mask platform) MT of programmable patterning device It remains substantially static, and while being moved or scanned to the substrate table WT, the figure of the radiation beam will be assigned Case projects on target part C.In this mode, impulse radiation source is generallyd use, and in each of the substrate table WT After secondary movement or between the continuous radiation pulse during scanning, the programmable patterning device is updated as needed. This operation mode can be easy to be applied to using programmable patterning device (for example, the programmable reflection of type as described above Lens array) maskless lithography art in.

In addition, lithographic equipment can be with two or more (such as two or more substrate tables, two or more More patterning device platforms and/or substrate table and platform without substrate) type.It, can be in such " multi-platform " device Additional is concurrently used, or can be while carrying out preliminary step on one or more platforms, it will be one or more Other for exposing.For example, Double tabletop lithographic equipment is described in 969,441 in United States Patent (USP) US5, by quoting it It is incorporated herein.

Figure 12 illustrates in greater detail equipment 1000, including source collector module SO, irradiation system IL and optical projection system PS. Source collector module SO, which is so constructed and arranged that, keeps vacuum environment in the encirclement structure 220 of source collector module SO.With Plasma source can be generated by electric discharge in the plasma 210 of transmitting EUV radiation to be formed.EUV radiation can pass through gas Or steam generation, such as xenon, lithium steam or tin steam, wherein forming high isothermal plasma 210 to emit in electromagnetic radiation Radiation in the EUV range of spectrum.Very high temperature is formed for example, by causing the electric discharge of the plasma at least partly ionized Plasma 210.For example, may require Xe, Li, Sn steam or any other suitable gas or steaming to efficiently generate radiation The partial pressure of the 10Pa of vapour.In one embodiment, the plasma for the tin (Sn) being excited is provided to generate EUV radiation.

The radiation emitted by high-temperature plasma 210 is from source chamber 211 via in the opening being optionally located in source chamber 211 Or gas barrier behind or contaminant trap 230 (being referred to as contamination barrier or foil trap in some cases) are passed It is delivered in collector chamber 212.Contaminant trap 230 may include channel design.Contaminant trap 230 can also include gas barrier Or the combination of gas barrier and channel design.The contaminant trap or contamination barrier 230 further shown herein is at least wrapped Channel design is included, as be known in the art.

Collector chamber 211 may include radiation collector CO, can be so-called grazing incidence collector.Radiation collector CO has upstream radiation collector side 251 and downstream radiation collector side 252.Radiation across collector CO can be reflected off Grating spectral filter 240 is opened to be focused on virtual source point IF along the optical axis indicated by dotted line ' O '.Virtual source point IF is commonly referred to as Intermediate focus, and the source collector module is arranged so that intermediate focus IF is located at the opening for surrounding structure 220 or its is attached Closely.Virtual source point IF is the picture for emitting the plasma 210 of radiation.

Then radiation may include being arranged to provide at patterning device MA across irradiation system IL, irradiation system IL The desired angle of radiation beam 21 is distributed and provides at patterning device MA the facet field of desired radiation intensity uniformity Reflector apparatus 22 and facet pupil reflector apparatus 24.In radiation beam 21 in the patterning device kept by support construction MT When reflecting at MA, patterned beam 26 is formed, and patterned beam 26 passes through optical projection system PS via reflecting element 28,30 It is imaged on the substrate W kept by substrate table WT.

It may be generally present elements more more than the element of diagram in illuminated optical apparatus unit IL and optical projection system PS. Grating spectral filter 240 can be optionally arranged, this depends on the type of lithographic equipment.Furthermore, it is possible in the presence of than showing in figure The more reflecting mirrors of reflecting mirror out, such as there may be the 1- in addition to the element being shown in FIG. 12 in optical projection system PS 6 additional reflecting elements.

Collector Optical devices CO, as shown in figure 12, be shown to be in figure with grazing incidence reflector 253,254 and 255 nido collector, only as an example of collector (or collector reflection mirror).Grazing incidence reflector 253,254 with And 255 axially and symmetrically be arranged around optical axis O, the collector Optical devices CO of the type is it is preferably expected that generate plasma with electric discharge Body source is used in combination, and it is commonly referred to as the source DPP that electric discharge, which generates plasma source,.Alternatively, source collector module SO can be LPP A part of radiating system.

Term " optical projection system " used herein can be broadly interpreted as encompassing any type of optical projection system, including folding Emitting, reflection-type, reflection-refraction type, magnetic type, electromagnetic type and electrostatic optical systems, or any combination thereof, as make It is that exposing radiation is suitble to or for being such as suitble to using immersion liquid or using the other factors of vacuum etc.

The lithographic equipment can also be this seed type: wherein at least part of substrate can be by with relatively high folding Liquid (such as water) covering of rate is penetrated, to fill the space between optical projection system and substrate.Immersion liquid applies also to The space between other spaces, such as mask and optical projection system in lithographic equipment.Immersion technique is for improving optical projection system What numerical aperture was well-known in the art.Term " submergence " used herein is not meant to that structure (such as substrate) must be soaked Enter into liquid, and mean onlys that in liquid in exposure process between optical projection system and the substrate.

Design disclosed herein can be used for simulating any device manufacturing processes for being related to lithographic equipment or mathematically right Any device manufacturing processes for being related to lithographic equipment are modeled, thereby increases and it is possible to can generate wavelength that size constantly becomes smaller The appearance of imaging technique is particularly useful.The existing technology used includes DUV (deep UV) photolithography, can 193nm wavelength is generated with ArF laser, it might even be possible to the wavelength of 157nm is generated with fluorine laser device.In addition, EUV lithography art can Generate the wavelength within the scope of 5-20nm.

Although design disclosed herein can be used for manufacturing device on substrate (such as silicon wafer), should manage Solution, disclosed design can be used together with the photolithographic imaging system of any other type, such as removing silicon wafer Except substrate on the photolithographic imaging system that is imaged.

Patterning device mentioned above includes or can form design layout.Can use CAD, (area of computer aided is set Meter) program generates design layout.The process is commonly referred to as EDA (electric design automation).Most of CAD programs follow One group of scheduled design rule, for generating Functional Design layout/patterning device.These rules are by processing and design limitation To set.For example, design rule defines that the interval between circuit devcie (grid, capacitor etc.) or interconnection line is allowed Degree, in order to ensure that circuit devcie or line will not interact in a manner of not being expected to.Design rule limitation is typically referred to as For " critical dimension " (CD).The critical dimension of circuit can be defined as minimum widith or two lines or two holes of line or hole Between minimum interval.Therefore, CD has determined the overall dimensions and density of designed circuit.Certainly, in IC manufacturing Target first is that faithfully reappearing original circuit design (via patterning device) on substrate.

The term " mask " used in this case or " patterning device " can be construed broadly to indicate can be with For being assigned to correspond to the patterning cross section for the pattern that will be generated in the target part of substrate for incident radiation beam General patterning device；Term " light valve " can be used for this situation.In addition to traditional mask (transmission-type or reflection Formula mask；Binary mask, phase shifting mask, mixed type mask etc.) except, the example of other patterning devices includes:

Array of programmable mirrors.One example of such device has viscoelastic control layer and reflecting surface Matrix-addressable surface.Basic principle based on such equipment is the addressed areas of (for example) reflecting surface by incident spoke Penetrate and be reflected into diffraction radiation, and unaddressed areas domain by reflecting incident radiation at non-diffraction radiation.It, can be with using suitable optical filter The non-diffraction radiation is filtered out from reflecting bundle, to only leave diffraction radiation later；In this way, the beam can be sought according to matrix The addressing-pattern on location surface and be patterned.Required matrix addressing can be carried out by using suitable electronic device.It closes It may refer to such as United States Patent (USP) No.5,296,891 and No.5,523,193 in more information of such reflection mirror array, It is incorporated them into herein by reference.

Programmable LCD array.Such example constructed is given in United States Patent (USP) No.5,229,872, is led to Reference is crossed to be incorporated into herein.

As noted, microlithography is the important step in the manufacture of device (such as integrated circuit), wherein in substrate The pattern of upper formation defines the function element of IC, microprocessor, memory chip etc..Similar photoetching technique is also used for Form flat-panel monitor, MEMS (MEMS) and other devices.

Printing, there is the process of the feature of size of the classical resolution limit less than lithographic projection apparatus usually to be claimed For low k₁Photolithography is based on resolution formula CD=k₁× λ/NA, wherein λ is used radiation wavelength (currently most of It is 248nm or 193nm in situation), NA is the numerical aperture of the projecting optical device in lithographic projection apparatus, and CD is " critical ruler It is very little " (usually printed minimum feature size) and k₁It is experience resolution factor.In general, k₁It is smaller, it is multiple on substrate Existing pattern (the similar shape and size for being the specific Electricity Functional of acquisition and performance by circuit designers and being designed) becomes more difficult. In order to overcome these difficulties, the complicated step that fine-tunes is applied to lithographic projection apparatus and/or design layout.These are for example The optimization of including but not limited to NA and optical coherence setting, the illumination scheme of customization, phase-shift pattern formed device use, Optical proximity correction (OPC, sometimes referred to as " optics and course corrections ") in design layout is generally defined as " differentiating Other methods etc. of rate enhancing technology (RET) ".

As an example, optical proximity correction solve the problems, such as be the design layout being projected onto figure The final size of picture and positioning by not with the size of the design layout on patterning device and positioning unanimously or not only only according to Rely the size and positioning in the design layout on patterning device.It would be recognized by those skilled in the art that especially in photoetching In art simulation/optimization situation, term " mask "/" patterning device " and " design layout " can be mutually general, this be because In simulating/optimizing in photolithography, what the patterning device of physics was not required to use, but can be with design layout come generation The patterning device of table physics.For the small characteristic size occurred in some design layouts and high characteristic density, give The position for determining the particular edge of feature will be influenced to a certain extent by other the present or absent of adjacent feature.These are adjacent The light of nearly small quantity of the effect due to being coupled to another feature from a feature and generate and/or (all by non-geometric optical effect Such as diffraction and interference) it generates.Similarly, kindred effect may be by usually baking (PEB) after the exposure after photolithography, resisting Diffusion and other chemical effects during losing agent development and etching generate.

In order to assist in ensuring that the projected image in design layout is consistent with the given demand of objective circuit design, need to make Kindred effect is predicted and compensated with complicated numerical model, the correction for being directed to design layout or predeformation.Article " Full- Chip Lithography Simulation and Design Analysis-How OPC Is Changing IC Design ", C.Spence, Proc.SPIE, Vol.5751, pp 1-14 (2005) provide current " based on model " optics The general introduction of proximity correction process.In typical high-end designs, almost each feature of design layout has some repair Change, to realize projected image for the high fidelity of target design.These modifications may include the position of marginal position or line width It moves or biases and the application of " auxiliary " feature, " auxiliary " feature are used to assist the projection of other feature.

Be not usually " accurate science " using OPC, but empirical iterative process, be not always able to compensate it is all can The kindred effect of energy.Therefore, OPC effect (such as design layout after application OPC and any other RET) should pass through Design review (check) (DR) is verified, that is, is simulated using the thorough full chip of the numerical process model by calibration, is set to minimize Meter defect is introduced into the probability in patterning device pattern.

OPC and full chip RET verifying can be based on such as example in the United States Patent (USP) Shen of Publication No. US20050076322 Please with entitled " Optimized Hardware and Software For Fast, Full Chip Simulation ", Y.Cao Numerical model system and method described in the article of et al., Proc.SPIE, Vol.5754,405 (2005).

A kind of adjustment of the global deviation of RET and design layout is related.Global deviation is the pattern in design layout and beats Calculate the difference of the pattern of printing on substrate.For example, the circular pattern of 25nm diameter can be straight by the 50nm in design layout The pattern of diameter is printed on substrate, or is printed onto substrate on by the pattern of 20nm diameter in design layout and with large dosage.

Other than the optimization to design layout or patterning device (such as OPC), irradiation source can also be optimised, or Person optimizes together with patterning device optimization or individually optimizes, and is dedicated to improving whole photoetching fidelity. Term " irradiation source " herein and " source " can be mutually general.It is well known that off-axis illumination, such as annular, quadrupole and Bipolar irradiation be it is attested for differentiate include fine structure (such as target signature) in patterning device side Formula.However, off-axis illumination source is usually that spatial image (AI) provides lower radiation by force when compared with traditional irradiation source Degree.Therefore, it is necessary to attempt to optimize irradiation source, to obtain the flat of optimization between finer resolution ratio and reduced radiation intensity Weighing apparatus.

For example, in entitled " the Optimum Mask and Source Patterns to Print A such as Rosenbluth Given Shape ", Journal of Microlithography, Microfabrication, Microsystems 1 (1), PP.13-20, in the article of (2002), it can be found that many irradiation source optimization methods.The source is subdivided into multiple regions, Each region corresponds to the specific region of pupil spectrum.Later, it is assumed that source distribution is uniform in each source region, and for Processing window optimizes the brightness in each region.In entitled " the Source Optimization for Image of Granik Fidelity and Throughput ", Journal of Microlithography, Microfabncation, Microsystems 3 (4), PP.509-522 in another example described in the article of (2004), are reviewed several existing Source optimization method, proposes the method based on illuminator pixel, this method by source optimization problem be converted into it is a series of it is non-negative most Small two multiply optimization.

For low k₁Photolithography, to the optimization in source and patterning device for ensuring the projection for critical circuit pattern Feasible processing window be highly useful.Some algorithms make irradiation be separated into independent source point and make patterning device The order of diffraction and the Kernel-based methods window metric (such as exposure latitude) being separated into spatial frequency domain are independently expressed with formula Cost function (its function for being defined as selected design variable), the processing window measurement can pass through optical imagery mould Type is predicted by source point intensity and the patterning device order of diffraction.Term " design variable " used herein means equipment Or one group of parameter of device manufacturing processes, such as lithographic equipment the adjustable parameter of user or user can pass through tune These whole parameters are come the picture characteristics that adjusts.It should be appreciated that (including source, pattern are formed any characteristic of device manufacturing processes These characteristics in device, projecting optical device and/or resist characteristic) it can be among design variable in optimization.Cost Function is usually the nonlinear function of design variable.Standard optimization techniques are for minimizing cost function later.

Source and patterning device (design layout) optimization method and system allow use cost function without constraint and Optimize source and patterning device simultaneously in practicable time quantum, in commonly assigned Publication No. WO2010/ It is described in 059954 PCT Patent Application, is hereby incorporated by reference in its entirety..

Another provenance and photomask optimization method and system are related to optimizing the source by adjusting the pixel in source, open Number to be described in 2010/0315614 U.S. Patent application, it is hereby incorporated by reference in its entirety..

It includes various types of optical systems that term " projecting optical device " as employed herein, which should be interpreted broadly into, System, for example including refractive optical device, reflection type optical device, aperture and refraction reflection type optical device.Term " projected light Learn device " it can also collectively or individually include according to these design classes for guiding, shaping or controlling projection beam of radiation The component that any one of type is operated.Term " projecting optical device " may include any light in lithographic projection apparatus Department of the Chinese Academy of Sciences's part is on which position in the optical path of lithographic projection apparatus but regardless of optical component.Projecting optical device can wrap It includes for being shaped before patterning device in radiation, the optical component of the radiation of adjustment and/or projection from source, and/ Or the optical component for shaping, adjusting and/or projecting the radiation after patterning device in radiation.Projection optics dress Setting usually does not include source and patterning device.

Although have been made above with specific reference to, in the case where the embodiment of the present invention is used for optical lithography, Such as the imprint lithography it will be noted that the embodiment of the present invention can be used in other applications, as long as and situation permit Perhaps, it is not limited to optical lithography.In imprint lithography, pattern in patterning device defines to be generated on substrate Pattern.The pattern of the patterning device can be printed onto the resist layer for being supplied to the substrate, be passed through on it Apply electromagnetic radiation, heat, pressure or combinations thereof solidify the resist.After resist solidification, the pattern It forms device to be removed from the resist, and leaves pattern in the resist.Therefore, using the lithographic equipment of stamping technique Generally include the template holding meanss for keeping impression block, the substrate table for keeping substrate and one or more causes Dynamic device, the actuator is used to cause the relative movement between substrate and impression block so that the pattern of impression block can be with It is stamped on the layer of substrate.

Following aspects can be used and further describe the present invention:

1. a kind of failure prediction methods that computer executes, for device manufacturing processes, the device manufacturing processes relate to And the production substrate handled by lithographic equipment, which comprises

Carry out train classification models using training group, the training group includes and the life as handled by the device manufacturing processes Produce the associated procedure parameter of substrate measured value or determine value and about under the process parameter values in the device The existing instruction of the handled production associated defect of substrate in manufacturing process；With

Output is generated from disaggregated model, indicates the prediction of the defect for substrate.

2. according to method described in aspect 1, including another training group is used to come train classification models, another training group Other measured value or determination including procedure parameter associated with the production substrate as handled by the device manufacturing processes Value and about with production substrate phase handled in the device manufacturing processes under the other value in the processing parameter The existing instruction of associated defect.

3. according to method described in aspect 2, wherein at least some of described other value is using measured value or determination It is generated after value train classification models.

4. the method according to aspect 2 or 3, wherein another training group includes the institute in addition to the other value It states measured value or determines at least part of value.

5. the method according to any one of aspect 1 to 4, further include based on as handled by the device manufacturing processes The other associated procedure parameter of production substrate other measured value or determination value be repeatedly trained.

6. the method according to any one of aspect 1 to 5 further includes the defect calculated using disaggregated model for substrate Probability.

7. according to method described in aspect 6, further includes: adjust the ginseng of the device manufacturing processes using the probability It number, the parameter of layout that is patterned on substrate or the parameter of the device manufacturing processes and is patterned on substrate Both parameters of layout.

8. the method according to any one of aspect 1 to 7, wherein the instruction existing for defect includes by light It learns measuring tool or operator's input or determines determined by yields data or electrical test data.

9. the method according to any one of aspect 1 to 8, wherein the instruction existing for defect includes by passing through Test the judgement that model or computation model carry out.

10. the method according to any one of aspect 1 to 9, wherein the instruction existing for defect include by The judgement that the user of lithographic equipment carries out.

11. the method according to any one of aspect 1 to 10, wherein the instruction existing for defect is included in By layout patterns on each tube core or each substrate of substrate after judgement.

12. the method according to any one of aspect 1 to 11, wherein disaggregated model is related to Luo Jisi recurrence, kernel sieve Ji Si recurrence, support vector machine or input vector machine.

13. the method according to any one of aspect 1 to 12, wherein the categorical measure of disaggregated model is 2.

14. according to method described in aspect 13, wherein the classification includes the presence of defect and being not present for defect.

15. the method according to any one of aspect 1 to 14, wherein the defect is selected from by constriction, line rollback, line It narrows, one or more kinds of defects in the group that critical dimension, overlap joint and bridge joint are constituted.

16. the method according to any one of aspect 1 to 15, wherein the parameter of the device manufacturing processes be selected from by The characteristic of the radiation source of lithographic equipment, the characteristic of the projecting optical device of lithographic equipment, dosage, focusing, resist characteristic, anti- It is one or more in the group that the characteristic of erosion agent development, the etching characteristic of the characteristic and substrate that bake after anti-aging drug are constituted Parameter.

17. the method according to any one of aspect 1 to 16 further includes that will be patterned on substrate using utilization The value of the procedure parameter of the parameter simulation of layout and existing for defect associated with the value of the simulation of procedure parameter Instruction carrys out train classification models.

18. the method according to any one of aspect 1 to 17 further includes being joined using the process as measured by measuring tool Several values carry out train classification models.

19. the method according to any one of aspect 1 to 18 further includes determining about relevant to the value of procedure parameter The existing instruction of defect.

20. the method according to any one of aspect 1 to 19 further includes the value of measurement or determination process parameter, described Value is selected from from the measured value of measuring tool, yields data or from one of value of lithographic equipment or more Value.

21. the method according to any one of aspect 1 to 20, wherein the device manufacturing processes are etching processes.

22. the method according to any one of aspect 1 to 20, wherein the device manufacturing processes include lithographic patterning Process.

23. a kind of method of train classification models, which comprises

The defect in the substrate or on substrate is predicted using disaggregated model, and the disaggregated model has as independent variable The procedure parameter of device manufacturing processes for photolithographic exposure substrate and/or lithographic equipment will be used to be provided to the figure on substrate The layout parameter of case；

Receive the existing letter about the measured value for being directed to procedure parameter and/or layout parameter or the defect for determining value Breath；With

Based on the defect predicted and about the measured value for being directed to procedure parameter and/or layout parameter or determine lacking for value Sunken existing information carrys out train classification models.

24. according to the method for aspect 23, wherein the information existing for defect includes by Optical measuring tool institute Multiple values of the procedure parameter of the device manufacturing processes of measurement.

25. the method according to aspect 23 or 24 further includes being based on manufacturing during device manufacturing processes from by device Data measured by multiple substrates handled by process execute prediction, reception and training repeatedly.

26. the method according to any one of aspect 23 to 25, further includes: using the output of disaggregated model to adjust State the parameter of device manufacturing processes, the parameter of the parameter for the layout being patterned on substrate or the device manufacturing processes and Both the parameters of layout being patterned on substrate.

27. the method according to any one of aspect 23 to 26, wherein disaggregated model is related to Luo Jisi recurrence, kernel sieve Ji Si recurrence, support vector machine or input vector machine.

28. one kind is performed by computer, generates disaggregated model to predict in the defects of device manufacturing processes Method, the device manufacturing processes are related to the production substrate as handled by lithographic equipment, and the method includes using training group Train classification models, the training group include the measurement of the procedure parameter of multiple substrates as handled by the device manufacturing processes Value determines value and the existing instruction about defect associated with the value of the processing parameter.

29. further including the defect predicted using disaggregated model in the substrate according to method described in aspect 28.

30. further including providing the estimation of the probability of the defect according to method described in aspect 29.

31. a kind of computer program product, including computer-readable medium, the computer-readable medium, which has, to be recorded in Instruction thereon, described instruction realize method described in any one of above-mentioned aspect when being executed as computer.

Above description is illustrative and not restrictive.It therefore, it should be apparent to those skilled in the art that can be not As made modification describedly in the case where the range of the attached claims.

Claims (14)

1. a kind of failure prediction methods performed by computer, for device manufacturing processes, the device manufacturing processes are related to The production substrate handled by lithographic equipment, which comprises

Carry out train classification models using training group, the training group includes serving as a contrast with the production as handled by the device manufacturing processes The measured value of the associated procedure parameter in bottom or determine value and about in the measured value of the procedure parameter or under determining value The existing instruction of the handled production associated defect of substrate in the device manufacturing processes；With

Output is generated from disaggregated model, the output indicates the prediction of the defect for substrate,

Wherein the method also includes based on associated with production substrate other as handled by the device manufacturing processes The other measured value of procedure parameter determines value repeatedly to carry out the training step.

2. according to the method described in claim 1, include carrying out train classification models using another training group, another training group Other measured value or determination including procedure parameter associated with the production substrate as handled by the device manufacturing processes Value and about with the other measured value in the procedure parameter or determine under value handled in the device manufacturing processes Production the associated defect of substrate existing instruction.

3. according to the method described in claim 2, wherein at least some of the other measured value or determining value are using The measured value determines that value train classification models are generated later.

4. described another according to the method in claim 2 or 3, wherein except the other measured value or in addition to determining value Training group further includes the measured value or at least part for determining value.

5. according to the method described in claim 1, further including the probability for calculating the defect for substrate using disaggregated model.

6. according to the method described in claim 5, further include: the ginseng of the device manufacturing processes is adjusted using the probability It number, the parameter of layout that is patterned on substrate or the parameter of the device manufacturing processes and is patterned on substrate Both parameters of layout.

7. according to the method described in claim 1, wherein the instruction existing for defect includes by Optical measuring tool Or operator inputs the judgement carried out or determines determined by yields data or electrical test data.

8. according to the method described in claim 1, wherein the instruction existing for defect includes by empirical model or meter Calculate model carry out judgement or by lithographic equipment user carry out judgement or by layout patterns in each of substrate The judgement carried out after on tube core or each substrate.

9. according to the method described in claim 1, wherein disaggregated model is related to Luo Jisi recurrence, kernel Luo Jisi is returned, is supported Vector machine or input vector machine.

10. according to the method described in claim 1, wherein the non-independent variable of disaggregated model is binary.

11. according to the method described in claim 10, wherein the non-independent variable includes the presence of defect and not depositing for defect ?.

12. according to the method described in claim 1, wherein the defect is selected from by constriction, line retract, line narrows, critical ruler The one or more kinds of defects in group that very little, overlap joint and bridge joint are constituted.

13. a kind of method of train classification models, which comprises

The defect in the substrate or on substrate is predicted using disaggregated model, and the disaggregated model has to be used for as independent variable The cloth of the procedure parameter of the device manufacturing processes of photolithographic exposure substrate and/or the pattern being provided to using lithographic equipment on substrate Office's parameter；

Receive the existing information about the measured value for being directed to procedure parameter and/or layout parameter or the defect for determining value；With

Based on the defect predicted and about the defect of the measured value for being directed to procedure parameter and/or layout parameter or determining value Existing information carrys out train classification models, wherein the train classification models include using training group train classification models, it is described Training group includes the measured value of the procedure parameter of multiple substrates as handled by the device manufacturing processes or determines value and close In the existing instruction of defect associated with the measured value of the procedure parameter or determining value；

Wherein the method also includes based on associated with production substrate other as handled by the device manufacturing processes The other measured value of procedure parameter determines value come the step of being repeatedly trained the disaggregated model.

14. it is a kind of it is performed by computer, generate disaggregated model in order to the method predicted in the defects of device manufacturing processes, The device manufacturing processes are related to the production substrate as handled by lithographic equipment, and the method includes using training group to train point Class model, the training group include the measured value of the procedure parameter of multiple substrates as handled by the device manufacturing processes or really Definite value and about with the measured value of the procedure parameter or determining the existing instruction of the associated defect of value,

Wherein the method also includes based on associated with production substrate other as handled by the device manufacturing processes The other measured value of procedure parameter determines value come the step of being repeatedly trained the disaggregated model.